Edge cooperative control method of industrial internet connector

Abstract

The invention relates to the technical field of industrial internet control, and discloses an edge cooperative control method of an industrial internet connector. The method comprises the following steps: acquiring operation data of a plurality of edge nodes, wherein the data comprises performance indexes of all connectors in each preset time period; comparing the performance index changes of each connector and all connectors in each edge node to obtain the performance change deviation of each connector; the performance index distribution range and the deviation are combined, and the performance change value of each connector is analyzed and obtained; obtaining the spatial distribution difference of each connector by comparing the performance distribution of each connector and other connectors with the performance difference of adjacent connectors; obtaining the collaboration degree of each edge node based on the performance change values and spatial distribution differences of all connectors; and finally, distributing a control strategy for the edge node according to the collaboration degree, and executing edge collaboration control. According to the method, the operation state of the connector can be comprehensively mastered, the node cooperation capability is accurately evaluated, and differential control is realized.

Description

Technical Field

[0001] This invention relates to the field of industrial internet control technology, specifically to an edge collaborative control method for industrial internet connectors. Background Technology

[0002] With the rapid development of Industrial Internet technology, the demand for connectivity between various industrial devices is increasing. As a core component for data transmission and command interaction between devices, connectors directly affect the operational efficiency of the entire Industrial Internet system through their operational stability and collaborative efficiency. Currently, Industrial Internet systems generally adopt a distributed architecture with multiple edge nodes, where each edge node deploys a varying number of connectors to undertake functions such as data acquisition, transmission, and control command forwarding within the region.

[0003] In actual operation, the industrial environment is complex and ever-changing. External factors such as temperature, humidity, and electromagnetic interference, as well as internal factors such as equipment load fluctuations and changes in data transmission volume, can all cause dynamic changes in connector performance. Due to differences in location, operating environment, and tasks, the performance variation patterns of connectors at different edge nodes vary, and there are implicit performance correlations between connectors. Focusing on the performance status of a single connector or edge node makes it difficult to fully grasp the overall operation of connectors in the entire system.

[0004] In existing technologies, connector control often employs centralized management or single-node independent control modes. In the centralized management mode, connector operating data from all edge nodes needs to be aggregated to a central control platform for analysis and decision-making. However, industrial sites generate massive amounts of data, and data transmission is prone to delays and packet loss, leading to untimely control command issuance and impacting system response speed. The single-node independent control mode formulates control strategies based solely on local connector operating data within its own node, neglecting the performance correlation and coordination requirements between connectors at different edge nodes. When connector performance in a certain area is abnormal, it's impossible to achieve overall system performance balance through coordinated adjustments of connectors at adjacent nodes, easily leading to the spread of local performance bottlenecks and reducing the operational stability and reliability of the entire industrial internet system.

[0005] Current technologies for evaluating connector performance often focus on monitoring the absolute values ​​of single performance indicators, such as data transmission rate and bit error rate, failing to fully consider the dynamic trends of performance indicators and the performance distribution differences between different connectors. When connector performance indicators are within normal threshold ranges but exhibit abnormal fluctuations, or when the performance differences between different connectors are too large, current technologies struggle to identify potential risks in advance and cannot adjust control strategies in a timely manner. This results in connectors operating in suboptimal conditions for extended periods, affecting data transmission quality, potentially shortening connector lifespan, and increasing the maintenance costs and failure risks of industrial internet systems. Summary of the Invention

[0006] The purpose of this invention is to provide an edge collaborative control method for industrial internet connectors to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides an edge collaborative control method for industrial internet connectors, the method comprising:

[0008] Acquire operational data from multiple edge nodes, with each data set containing performance metrics for all connectors within a preset time period;

[0009] By comparing the changes in the performance indicators of each connector in each edge node with those of all connectors in all edge nodes, the performance variation deviation of each connector in each edge node can be obtained.

[0010] Analyze the distribution range of the performance indicators of each connector in each edge node and the deviation of the performance change, and obtain the performance change value of each connector in each edge node;

[0011] By comparing the performance distribution of each connector in each edge node with all other connectors, and the performance difference between each connector and its adjacent connectors at its location, the spatial distribution difference of each connector in each edge node is obtained.

[0012] By combining the performance variation values ​​and spatial distribution differences of all connectors in each edge node, the degree of coordination of each edge node is obtained;

[0013] Control strategies are assigned to all edge nodes based on the degree of collaboration, and edge collaborative control is performed on the edge nodes after the control strategies are assigned.

[0014] Preferably, the process for obtaining the performance change deviation is as follows:

[0015] All performance index data of each connector in each edge node are arranged in time sequence to form performance sequences. A filtering algorithm is used to process all performance index data in each performance sequence to obtain a smooth curve for each performance sequence.

[0016] Obtain the curvature calculation method for each smooth curve, and calculate the integral result of each curvature calculation method over the entire time range;

[0017] Calculate the average of the integral results for all connectors in all edge nodes;

[0018] The performance variation deviation is obtained by comparing the integral result of each connector in each edge node with the average value.

[0019] Preferably, the performance variation deviation is specifically the difference between the integral result of each connector in each edge node and the average value.

[0020] Preferably, the process for obtaining the performance change value is as follows:

[0021] Calculate the range of all performance metrics for each connector in each edge node;

[0022] The performance change values ​​are positively correlated with the range and the performance change deviation, respectively.

[0023] Preferably, the process for obtaining the spatial distribution difference is as follows:

[0024] Calculate the average performance index data of each row of connectors in each edge node at each acquisition time. Record the difference between the average performance index data of each row of connectors in each edge node and the row above it as the performance value difference. Accumulate the performance value differences of each connector in each edge node at all acquisition times to obtain the total performance difference value.

[0025] Obtain the extreme points of the smooth curve of each connector, arrange the occurrence times of the extreme points of all connectors in each edge node into a time series according to the time sequence, calculate the time interval between adjacent times in the time series to form a time difference series, and calculate the average interval value of the time difference series.

[0026] The spatial distribution difference is obtained by processing the ratio of the total performance difference value to the average interval value using a normalization function.

[0027] Preferably, the process of obtaining the degree of synergy is as follows:

[0028] Calculate the weighted sum of the performance variation values ​​of each connector in each edge node and the spatial distribution differences;

[0029] The degree of synergy is the sum of the weighted sums of the values ​​of all connectors in each edge node.

[0030] Preferably, the process of assigning control strategies to all edge nodes based on the degree of cooperation is as follows:

[0031] Based on the degree of collaboration of each edge node, an initial cluster center is selected using a genetic algorithm. Then, a clustering algorithm is used to cluster all edge nodes based on the initial cluster center. The clustering results are used to distinguish between edge nodes with high collaboration and edge nodes with low collaboration. Different control strategies are assigned to edge nodes with high collaboration and edge nodes with low collaboration. The control strategy for edge nodes with high collaboration is the first preset strategy, and the control strategy for edge nodes with low collaboration is the second preset strategy.

[0032] Preferably, the edge node performing edge collaborative control based on the allocation control strategy includes:

[0033] All high-cooperation edge nodes are assigned to the priority execution group and the standard execution group according to a preset ratio, and low-cooperation edge nodes are assigned to the priority execution group and the standard execution group according to a preset ratio;

[0034] Edge collaborative control operations are performed using the priority execution group and the standard execution group.

[0035] Preferably, the method further includes:

[0036] The process of acquiring historical operating data for each edge node and correcting the performance change deviation based on the historical operating data.

[0037] Preferably, the method further includes:

[0038] The process of acquiring network topology data for each edge node and correcting the spatial distribution differences based on the network topology data.

[0039] Compared with the prior art, the beneficial effects of the present invention are:

[0040] By comprehensively acquiring the performance indicators of all connectors across multiple edge nodes within preset time periods, this approach overcomes the limitations of traditional data acquisition methods. It moves beyond focusing solely on localized data from a single node or a subset of connectors, incorporating the operational data of all connectors throughout the entire system into the analysis. This provides a more comprehensive and complete data foundation for subsequent performance evaluation and collaborative control. Based on multi-dimensional and full-range operational data, it more accurately reflects the performance changes of each connector across different time dimensions, avoiding performance evaluation biases caused by data limitations and making the judgment of connector operational status more accurate and reliable.

[0041] In the performance deviation acquisition stage, by comparing the performance index changes of each connector in each edge node with those of all connectors in all edge nodes, a correlation between individual and overall performance changes is established. This allows for a clear identification of the degree of deviation of a single connector's performance change relative to the overall system performance change. This comparative analysis method not only focuses on the absolute changes in the connector's own performance but also emphasizes its relative trend within the context of the overall system performance. This helps to promptly identify connectors whose absolute performance index values ​​are within the normal range but whose trends are abnormal, allowing for early detection of potential performance risks and providing a more sufficient time window for subsequent control strategy adjustments.

[0042] In acquiring performance change values, a comprehensive analysis combining the distribution range of performance indicators and performance change deviations, rather than relying solely on a single parameter, allows for a more holistic assessment of the dynamic characteristics of connector performance. The distribution range of performance indicators reflects the fluctuation range of connector performance within a certain period, while performance change deviations reflect its deviation relative to the overall picture. Combining these two aspects allows for a more accurate quantification of the actual degree of change in connector performance, avoiding inaccurate performance evaluations caused by focusing only on the distribution range while ignoring individual deviations, or focusing only on deviations while ignoring the overall distribution pattern. This ensures that the obtained performance change values ​​more objectively reflect the true operating status of the connector.

[0043] By comparing the performance distribution of each connector in each edge node with all other connectors, and the performance differences between each connector and its adjacent connectors, spatial distribution differences are obtained, fully considering the spatial location correlation of connectors in the system and the overall performance distribution characteristics. This analysis method can not only identify the position of a single connector in the overall system performance distribution and determine whether it is in a state of performance imbalance, but also focus on the performance synergy between adjacent connectors, discovering potential performance interference problems due to spatial proximity. For example, when the performance difference between a connector and its adjacent connectors is too large, potential local performance inconsistencies can be detected in a timely manner, providing a basis for subsequent regional performance balance through coordinated control.

[0044] By combining the performance variations and spatial distribution differences of all connectors, the coordination degree of each edge node is obtained. This organically integrates the individual performance characteristics of connectors with the overall coordination state of the nodes, achieving a leap from individual performance evaluation to node coordination state evaluation. The coordination degree can intuitively reflect the collaborative operation capability of each edge node in the entire system, clarifying the collaborative advantages and disadvantages between different edge nodes. This avoids the drawback of traditional control methods that only focus on the individual performance of nodes while ignoring the collaborative relationship between nodes, providing a key basis for subsequently formulating differentiated control strategies for different edge nodes.

[0045] By assigning control strategies to all edge nodes based on their degree of collaboration and executing edge collaborative control, precise matching and efficient execution of control strategies can be achieved. For edge nodes with varying degrees of collaboration, appropriate control strategies are formulated and assigned, avoiding the "one-size-fits-all" drawbacks of traditional centralized control and overcoming the lack of collaboration in independent control of a single node. Through edge collaborative control, each edge node can dynamically adjust according to its own degree of collaboration and the status of adjacent nodes. When the performance of a node connector malfunctions, the performance gap can be quickly compensated for through the collaborative cooperation of adjacent nodes, achieving overall system performance balance and stability. This reduces the risk of system failures caused by local performance issues, while ensuring that each connector is always in a relatively optimal operating state, reducing maintenance costs, extending equipment lifespan, and improving the overall operational efficiency and reliability of the industrial internet system. Attached Figure Description

[0046] Figure 1 This is a schematic diagram illustrating the working principle of the edge collaborative control method for the industrial internet connector described in this invention.

[0047] Figure 2 Flowchart for obtaining performance variation deviations;

[0048] Figure 3 A flowchart for obtaining spatial distribution differences. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] Please see Figure 1 This invention provides an edge collaborative control method for industrial internet connectors. The method includes: acquiring operational data from multiple edge nodes, each operational data containing performance indicators of all connectors within a preset time period; obtaining performance change deviations of each connector in each edge node by comparing the changes in performance indicators of each connector in each edge node with those of all connectors in all edge nodes; analyzing the distribution range of performance indicators of each connector in each edge node and the performance change deviations to obtain performance change values ​​of each connector in each edge node; obtaining spatial distribution differences of each connector in each edge node by comparing the performance distribution of each connector in each edge node with that of all other connectors, and the performance differences between each connector and its adjacent connectors; obtaining the degree of collaboration of each edge node by combining the performance change values ​​and spatial distribution differences of all connectors in each edge node; assigning control strategies to all edge nodes based on the degree of collaboration; and performing edge collaborative control based on the edge nodes after assigning control strategies.

[0051] Example 1: See Figure 2In the implementation of the edge collaborative control method for industrial internet connectors, acquiring performance variation deviations is a fundamental and crucial step, relying on in-depth processing and analysis of edge node operational data. Acquiring operational data from multiple edge nodes is the starting point. This data is continuously collected at preset time intervals. Each edge node contains several connectors, and the performance indicators of each connector, such as transmission latency, data throughput, and error rate, are recorded in timestamp order and encapsulated into data packets for transmission to the central processing unit. The integrity of the data acquisition is ensured through a verification mechanism; any lost or abnormal data frames trigger retransmission requests, ensuring the reliability of the original data. All performance indicator data of each connector in each edge node are arranged chronologically to form a performance sequence. This process is automatically completed in the data processing module. The length of the performance sequence depends on the preset time period division, typically forming a discrete time series in minutes or hours. Each performance sequence represents the performance evolution trajectory of a connector within a specific time period. Data points in the sequence are time-stamped for easy time series analysis. The sequence construction follows time alignment principles, ensuring comparability of data between different connectors.

[0052] A filtering algorithm is used to process all performance index data in each performance sequence. The choice of filtering algorithm depends on the data characteristics; moving average filtering or Gaussian filtering is widely used to smooth short-term fluctuations and suppress random noise. The setting of the filtering window size requires a trade-off between smoothing effect and response speed. An excessively large window may lead to the loss of detailed information, while an excessively small window will result in poor filtering. After filtering, each performance sequence is converted into a smooth curve, which retains the overall trend of performance index changes while eliminating most high-frequency interference components. Obtaining the curvature calculation method for each smooth curve involves the application of differential geometry concepts. Curvature is used to quantify the degree of curvature at each point, reflecting the severity of performance changes. The calculation is approximated by numerical differentiation on the curve. The first and second derivatives at each point are calculated using the difference method and substituted into the curvature formula to obtain the curvature distribution over time. Points with larger curvature values ​​correspond to regions of rapid change in performance indexes, which may indicate significant changes in connector state. The integral result of each curvature calculation method is calculated over the entire time range. This integration operation performs a definite integral on the curvature-time function, covering the entire preset time interval, and the integral result is a scalar value. It characterizes the cumulative fluctuation intensity of connector performance over a given time period. A larger integral value indicates more frequent or more drastic performance changes, and this value comprehensively reflects the complexity of the connector's dynamic behavior. Calculating the average of the integral results for all connectors across all edge nodes requires aggregating data from all edge nodes. First, the integral results for each connector are collected, and a general sample set is constructed. The average is calculated using an arithmetic mean method. This average represents the baseline level of performance fluctuation intensity for all connectors in the entire system, serving as a reference standard for subsequent comparisons. The calculation period for the average can be kept consistent with the data acquisition period to achieve dynamic updates.

[0053] Performance variation deviation is obtained by comparing the integral results of each connector in each edge node with the average value. The comparison operation uses subtraction to calculate the absolute difference, which quantifies the degree of deviation of the performance fluctuation of a single connector from the system average level. A positive deviation indicates that the performance fluctuation of the connector is higher than the average level, while a negative deviation indicates that it is lower than the average level. The magnitude of the deviation value directly reflects the degree of abnormality in the dynamic performance characteristics of the connector. Specifically, performance variation deviation is the difference between the integral result of each connector in each edge node and the average value. This difference can be treated as a signed value to retain the deviation direction information, or the absolute value can be taken to focus on the deviation magnitude. This deviation value is stored in a dedicated data structure and associated with the corresponding connector identifier for easy retrieval and use later. The deviation calculation process is executed periodically to ensure the timeliness of the data, and the system dynamically updates the deviation value based on new collected data. The entire process of obtaining performance variation deviation is implemented through a dedicated calculation module, which includes sub-modules such as data preprocessing, filtering, curvature calculation, integration, and average value comparison. Each sub-module adopts a pipelined processing method to improve computational efficiency. Key parameters involved in the calculation process, such as the filter window size and integration interval range, can be adjusted through configuration files, enabling the system to adapt to the needs of different application scenarios. The accuracy and reliability of the calculation results are ensured through multiple verification mechanisms, including data range checks and outlier handling procedures. The calculation of performance variation deviations provides important input for subsequent analysis of performance variation values ​​and synergy, allowing the system to extract characteristic quantities reflecting the dynamic characteristics of connectors from a large amount of operational data. These characteristic quantities overcome the limitations of relying solely on instantaneous index values ​​for evaluation. By focusing on the cumulative effect and relative differences of performance variations, the system can more accurately identify connectors whose behavior patterns deviate from the norm, providing data support for the formulation of targeted control strategies. This method enhances the refinement and intelligence of connector management in the Industrial Internet.

[0054] Example 2: See Figure 3The acquisition of performance change values ​​is based on a comprehensive analysis of the distribution range of connector performance indicators and existing performance change deviations. Calculating the range of all performance indicator data for each connector in each edge node is the first step. The range is calculated individually for each connector. The maximum and minimum values ​​are identified from the performance indicator data collected at all times within a preset time period; the arithmetic difference between these two values ​​is the range. This value simply and intuitively depicts the absolute fluctuation range of connector performance within this time period. A larger range indicates that the connector's performance is more unstable and the fluctuation amplitude is more significant. The performance change value is defined as a quantity that is positively correlated with both the range and the performance change deviation. This positive correlation is usually achieved through linear combination or nonlinear function mapping. For example, a weighted summation method can be used, multiplying the range and performance change deviation by corresponding positive weighting coefficients and then adding them together. The specific values ​​of the weighting coefficients are preset according to the different levels of concern regarding the fluctuation amplitude and deviation degree in the actual application scenario. Another approach is to multiply the two together and then multiply by a scaling factor to amplify or reduce their combined effect. Regardless of the mathematical form used, the core objective is to construct a comprehensive indicator that can simultaneously reflect the connector's own fluctuation amplitude and its deviation from the overall system fluctuation.

[0055] Obtaining spatial distribution differences requires analyzing the performance distribution relationship between connectors from a spatial perspective, calculating the average performance index data of each row of connectors in each edge node at each acquisition time. Connectors in an edge node are typically divided into several rows based on their physical installation location. For each specific acquisition time, the arithmetic mean of the performance index data of all connectors in each row within the node is calculated, thus obtaining the average performance value of that row at that time. These averages are organized into a two-dimensional data structure according to the sort number and acquisition time, facilitating subsequent spatial comparative analysis. The difference in the average performance index data between the row containing each connector in each edge node and the row above it is recorded as the performance value difference, which is usually calculated using absolute differences. For a specific connector, at a certain acquisition time, the difference between the average performance value of its row and the average performance value of its adjacent row is the performance value difference of that connector at that moment. This difference reflects the spatial performance inconsistency between the connector and its neighboring rows. A large performance value difference indicates that there may be performance imbalance or propagation anomalies between the row containing the connector and the row above it.

[0056] The total performance difference value is obtained by summing the performance differences of each connector in each edge node at all acquisition times. This summation operation is the sum of the performance differences of the same connector at different acquisition times. This total performance difference value is a cumulative quantity that comprehensively reflects the overall degree of performance inconsistency between the connector's row and the row above it throughout the entire time period. The larger the total performance difference value, the more obvious the cumulative effect of the connector's performance anomaly in the spatial dimension. The extreme points of the smoothed curves of each connector are obtained. Extreme points include local maxima and local minima, which correspond to key moments when the trend of performance index changes reverses. Extreme points are identified by examining the change in the sign of the first derivative of the smoothed curve or by directly comparing the numerical values. Each extreme point corresponds to a specific occurrence time, recording the time position where the connector's performance changes significantly. The occurrence times of the extreme points of all connectors in each edge node are arranged chronologically to form a time series. This series contains information on the time points when the performance of all connectors in this edge node changes significantly. The time points in the time series are arranged in chronological order and may contain repeated timestamps because multiple connectors may experience extreme points at the same time. The series reflects the overall temporal distribution characteristics of the performance fluctuations of the entire node.

[0057] The time intervals between adjacent moments in the time series are calculated to form a time difference sequence, which is obtained by subtracting the previous moment from the subsequent moment. All these time intervals are arranged sequentially to form the time difference sequence, which characterizes the temporal density distribution of node performance fluctuation events. Shorter time intervals indicate more frequent performance fluctuation events. The average interval of the time difference sequence is calculated using the arithmetic mean method, by adding all adjacent time intervals and dividing by the number of intervals. This average interval reflects the average frequency of node performance fluctuation events; a smaller average interval indicates more concentrated performance fluctuation events and an unstable overall node performance. Spatial distribution difference is obtained by processing the ratio of the total performance difference to the average interval using a normalization function. This ratio is calculated by dividing the total performance difference by the average interval, and it integrates information from both spatial performance differences and temporal fluctuation frequency. The normalization function typically uses a maximum-minimum normalization method to linearly transform the ratio to between zero and one, or a logarithmic function to compress the numerical range, ensuring that the final spatial distribution difference is within a standardized numerical range, facilitating subsequent comprehensive comparison with other indicators.

[0058] Example 3; Obtaining the degree of synergy requires combining the performance variation value and spatial distribution difference of each connector, and calculating the weighted sum of the performance variation value and spatial distribution difference of each connector in each edge node. This calculation process is performed independently for each connector. The performance variation value reflects the dynamic fluctuation characteristics of the connector's own performance, while the spatial distribution difference reflects its correlation characteristics with surrounding connectors in the spatial dimension. The two are combined linearly with weights to form a comprehensive evaluation value. The mathematical expression of the weighted sum is:

[0059] S ij =α·V ij +β·D ij

[0060] Wherein: S ij V represents the weighted sum of the j-th connectors in the i-th edge node. ij D represents the performance variation value of the connector. ij This represents the spatial distribution difference value. α and β are pre-defined weighting coefficients, satisfying the constraint α+β=1. The specific values ​​of the weighting coefficients depend on the degree of attention paid to performance fluctuations and spatial differences in the actual application scenario. This weighted sum comprehensively reflects the degree of anomaly of a single connector in both time and space dimensions; a higher value indicates that the connector's behavior pattern requires more attention.

[0061] Collaboration degree is defined as the sum of the weighted sums of the values ​​of all connectors in each edge node. This calculation is achieved by accumulating the S-values ​​of all connectors within the node. The expression for the summation is:

[0062]

[0063] Where: C i This represents the coordination degree of the i-th edge node, where n represents the total number of connectors in that node. This coordination degree reflects the overall operational coordination status of the edge nodes. A higher value indicates that there are multiple connectors with abnormal behavior within the node, resulting in poor overall coordination performance; a lower value indicates that the connectors within the node operate in a relatively consistent manner, resulting in good coordination performance.

[0064] When assigning control strategies to all edge nodes based on their synergy, the first step is to select initial cluster centers using a genetic algorithm based on the synergy of each edge node. The genetic algorithm uses the synergy value as the fitness function, optimizing the selection of cluster centers by simulating the natural selection process. During the algorithm initialization phase, multiple candidate cluster center sets are randomly generated, each containing a preset number of cluster centers representing possible node classification schemes. Selection is based on fitness values; the synergy value is converted into selection probabilities, with individuals with higher fitness having a greater chance of being selected for the next generation. Crossover generates new solutions by exchanging cluster center information between different individuals, while mutation randomly changes the values ​​of some cluster centers to increase population diversity. After multiple iterations, the algorithm outputs optimized initial cluster center positions, which better represent the actual distribution characteristics of the edge nodes. A clustering algorithm based on the initial cluster centers optimized by the genetic algorithm is then used to perform cluster analysis on all edge nodes. The clustering algorithm typically chosen is K-means or its improved variants. The algorithm represents each edge node as a data point, characterized by its coherence score. The clustering process assigns each node to the nearest cluster center based on proximity. After clustering, the clusters containing nodes with high coherence scores are designated as high-coherence edge node groups, and those containing nodes with low coherence scores are designated as low-coherence edge node groups. This classification method allows the system to identify anomalous node groups requiring special attention from a large number of edge nodes. Different control strategies are assigned to the identified high-coherence and low-coherence edge nodes. The control strategy for high-coherence edge nodes adopts a first preset strategy. This strategy typically includes relatively lenient management measures, such as reducing monitoring sampling frequency, extending status check intervals, and lowering data transmission priority. These measures help reduce system load and improve overall operating efficiency. The control strategy for low-coordination edge nodes adopts a second preset strategy, which includes stricter management measures such as increasing real-time monitoring frequency, strengthening data verification mechanisms, increasing task scheduling priority, and implementing rapid failover. These measures aim to enhance the monitoring and management of abnormal nodes and prevent local problems from spreading and affecting the entire system. Specific parameter settings for both strategies can be adjusted through configuration files to adapt to the actual needs of different application scenarios. The entire process of acquiring coordination level and allocating control strategies is implemented through a dedicated decision module, which includes sub-functional units such as data preprocessing, weighted calculation, cluster analysis, and strategy generation. The module adopts a modular design concept, with data exchange between functional units through standard interfaces, facilitating system maintenance and functional expansion.Intermediate results and final decision data generated during the calculation process are persistently stored in the database for subsequent analysis and query. At the same time, the system provides a visual interface to display the distribution of the degree of collaboration of each edge node and the status of strategy allocation, so that managers can intuitively understand the system operation status.

[0065] Example 4: When edge nodes perform edge collaborative control based on the allocation control strategy, the nodes need to be grouped according to the collaboration degree classification results. Assume an industrial internet system contains 200 edge nodes, of which 120 are high-collaboration nodes and 80 are low-collaboration nodes. The system's preset ratio is set to 30% of high-collaboration nodes in the priority execution group and 10% of low-collaboration nodes in the priority execution group. According to this ratio, 36 high-collaboration nodes are assigned to the priority execution group, and 84 nodes are assigned to the standard execution group; 8 low-collaboration nodes are assigned to the priority execution group, and 72 nodes are assigned to the standard execution group. The grouping operation is automatically completed by the node management module. This module calculates the number of nodes in each group according to the preset ratio, and then uses a random sampling algorithm to select a specified number of nodes from the corresponding category's node set and assign them to the priority execution group. The remaining nodes are automatically assigned to the standard execution group. During the grouping process, the physical distribution of nodes and network topology relationships need to be considered to avoid assigning multiple nodes in the same area to the same execution group. The system employs a geographic location-based uniform sampling algorithm to ensure that an appropriate number of nodes in each physical area are included in the priority execution group. This distribution method helps improve the system's ability to cope with regional failures. After grouping, the system generates a node grouping configuration table, recording information such as the unique identifier, coordination level, execution group, and execution status of each node. This configuration table is synchronized to all relevant control units and monitoring nodes to ensure that the entire system has a unified understanding of the node grouping situation (see Table 1).

[0066] Table 1: Edge Node Grouping Configuration

[0067] Node ID Coordination Level Execution Group Physical area Status indicator EN-045 high Priority Execution Group Area A active EN-078 high Standards Implementation Group Area B standby EN-123 Low Priority Execution Group Area C active EN-156 Low Standards Implementation Group Region D Under maintenance EN-189 high Priority Execution Group Area A active EN-202 Low Standards Implementation Group Area B active

[0068] When performing edge collaborative control operations using a priority execution group and a standard execution group, the two groups of nodes undertake different tasks and have different performance requirements. The priority execution group handles control commands and critical data transmission tasks with high real-time requirements, such as real-time device status monitoring, emergency alarm handling, and issuing high-priority control commands. These tasks require low transmission latency and high reliability; therefore, nodes in the priority execution group use dedicated communication channels and priority scheduling strategies to ensure timely processing of critical tasks. The standard execution group is mainly responsible for routine tasks such as batch data processing, historical log uploading, and non-real-time configuration updates. These tasks have relatively lower real-time requirements but involve large data volumes and require strong throughput. Collaboration between the two groups of nodes is achieved through message middleware for data synchronization and status coordination. The system uses a publish-subscribe model for inter-group communication. Priority execution group nodes publish status updates and event notifications in real time, while standard execution group nodes subscribe to relevant topics and respond. Simultaneously, standard execution group nodes periodically publish resource utilization and workload information for priority execution group nodes to reference when distributing tasks. This communication mechanism ensures that the two groups of nodes can understand each other's status in a timely manner, enabling coordinated work. The message middleware is deployed in a distributed architecture to avoid single points of failure affecting inter-group communication. Message transmission uses an acknowledgment mechanism to ensure reliability, and important messages are set up with a retransmission mechanism to prevent loss.

[0069] When performing edge collaborative control operations, the system selects an appropriate execution group for processing based on the task type and urgency. For real-time control commands, the task scheduler prioritizes assigning tasks to nodes in the priority execution group, while also considering the current load and network status of the nodes to select the most suitable node for execution. If a node in the priority execution group is overloaded or malfunctions, the task can be automatically degraded and assigned to a better-performing node in the standard execution group. This degradation mechanism ensures that the system can maintain basic operation even under abnormal circumstances. For batch processing tasks, the system primarily uses nodes in the standard execution group, but when resources in the standard execution group are insufficient, idle nodes in the priority execution group can be temporarily called upon to assist in processing. During task execution, the monitoring module continuously tracks the execution status and performance indicators of each node, collecting data such as response time, processing success rate, and resource utilization. This monitoring data is fed back to the control center in real time to evaluate the effectiveness of the current grouping strategy. If the overall performance of an execution group is found to be substandard or an abnormal situation occurs, the system can dynamically adjust the grouping ratio or reallocate nodes within the group. For example, when network conditions deteriorate, the proportion of priority execution groups can be temporarily increased to include more nodes in the priority protection scope; when the system is running stably, the proportion of priority execution groups can be appropriately reduced to save resource consumption. Node grouping configuration supports dynamic update functionality, allowing administrators to manually adjust grouping proportions or specify groups for specific nodes based on system operating conditions. All configuration changes are controlled through a version management mechanism; each modification generates a new configuration version and records the reason for the change and a timestamp, facilitating subsequent auditing and problem tracing. The system provides a configuration simulation function, allowing administrators to test the impact of new configurations on system performance before formal application, avoiding performance degradation caused by improper adjustments. Fault isolation and fault tolerance between execution groups are naturally achieved through the grouping mechanism; a node failure in one execution group will not directly affect the normal operation of another execution group. Priority execution group nodes employ redundant deployment and rapid migration mechanisms; when a node failure is detected, its load is automatically transferred to other normal nodes within the group; standard execution group nodes employ load balancing and task retry mechanisms to ensure the final completion of batch tasks. This grouping design improves system reliability and availability, ensuring the continuous and stable operation of critical business operations.

[0070] Example 5: Acquiring historical operational data for each edge node is a fundamental step in the implementation process. This historical data covers complete operational records from the past six months, including hourly performance metrics collected for each connector: transmission latency, packet success rate, and bandwidth utilization. This data is stored in a distributed database in time-series format, with each data point containing a timestamp, node identifier, connector identifier, and corresponding performance metric value. The historical data undergoes a preprocessing process, including data cleaning to remove obvious outliers, data alignment to ensure consistency across different sources on the timeline, and data normalization to convert metrics of different dimensions to the same numerical range. This preprocessed historical data forms the training dataset for subsequent model training and algorithm optimization.

[0071] The process of obtaining performance change deviations based on historical operating data employs a supervised learning method. Performance index sequences from historical data are used as input features, and observed connector anomalies are used as labels. The training process utilizes a Long Short-Term Memory (LSTM) neural network model, which captures the long-term dependencies of performance index changes over time. The model learns the complex mapping between the performance index sequence and the final deviation value, adjusting network parameters through backpropagation to minimize the difference between predicted and actual deviations. The trained model can output more accurate estimates of performance change deviations based on real-time input performance index sequences. These estimates comprehensively consider historical operating patterns and are more closely aligned with actual operating conditions than the original algorithm. The corrected performance change deviation acquisition process incorporates context awareness. The system considers not only performance data within the current time window but also operating patterns from the same historical time periods. For example, for industrial equipment with periodic operating characteristics, the system identifies the periodic pattern and adjusts the expected value range accordingly in deviation calculations, avoiding misclassifying normal periodic fluctuations as abnormal deviations. This context-aware mechanism is achieved through an attention mechanism, where the model automatically learns which historical periods are most relevant to the current moment and assigns them higher weights.

[0072] Obtaining network topology data for each edge node requires collecting information from multiple data sources, including network device configuration management systems, link status information in SDN controllers, and actual connectivity test results between nodes. Topology data includes physical connections between connectors, such as cable connection ports and switch configuration information; data transmission paths, such as OSPF routing tables and BGP peering relationships; and distances between nodes, including physical distance and network hop count. This data is integrated into a unified topology model, represented by a graph structure, where vertices represent network devices or connectors, edges represent connections, and edge weights represent distance or cost metrics. The process of correcting spatial distribution differences based on network topology data introduces a topology weight factor, which reflects the closeness of connectors within the network topology. The topology weight factor is calculated based on various factors, including physical connection distance, data transmission hop count, link bandwidth, and historical communication quality. Connector pairs with closer physical distances, fewer hop counts, higher bandwidth, and stable communication quality receive higher weight factors, indicating they are more closely connected in the topology. The weight factor is used to adjust the calculation of performance value differences; performance differences between connectors with high closeness are given greater importance because topologically close connectors should theoretically have more similar performance. The correction process also considers the characteristics of network paths; bottleneck links and redundant paths in data transmission paths both affect the calculation of spatial distribution differences. For connectors sharing bottleneck links, the system expects their performance indicators to exhibit similar variation patterns, and this correlation is incorporated into the difference calculation model. The model identifies potential bottleneck links by analyzing historical traffic data and adjusts the interpretation of performance differences between these connectors accordingly. The existence of redundant paths may cause performance divergence between connectors; the system identifies this situation and adjusts the expected range of differences accordingly.

[0073] Dynamic changes in network topology are also taken into account. The system periodically re-collects topology data and updates the topology model. When a change in the topology is detected, such as a link failure, the addition or removal of equipment, the system automatically triggers the recalculation of topology weight factors and gradually adjusts them to the new weight values, avoiding drastic fluctuations in spatial distribution difference calculations caused by sudden changes. This gradual adjustment mechanism ensures that the system can adapt to changes in network topology while maintaining the stability of the calculation results. The combination of historical operating data and network topology data is achieved through a multimodal learning framework, which processes both time-series data and graph structure data. The time-series branch processes historical performance index data, while the graph neural network branch processes network topology data. The outputs of the two branches are combined in a fusion layer to jointly contribute to the final spatial distribution difference calculation. This multimodal approach can capture the temporal dynamics of performance data and the spatial constraints of topological relationships, producing more accurate and robust estimates of spatial distribution difference. The correction algorithm is implemented using an online learning approach. The system continuously collects new operating data and uses it for model updates. The online learning algorithm employs a mini-batch gradient descent strategy, periodically updating model parameters with the latest data, enabling the model to adapt to slow changes in system operating conditions. Meanwhile, the model update process incorporates stability constraints to prevent excessive parameter adjustments from causing oscillations in the calculation results. This design allows the system to gradually adapt to changes in the operating environment while maintaining stability. The entire correction process is executed through a dedicated algorithm engine, which includes a data acquisition interface, a preprocessing module, a model training unit, and real-time computing components. The engine adopts a microservice architecture, with components interacting through lightweight communication mechanisms, supporting horizontal scaling to handle large-scale edge node clusters. The calculation results are pushed to the collaborative control engine in real time for subsequent collaborative degree calculations and control strategy generation, forming a complete control closed loop. The system provides a management interface to display the running status and performance evaluation metrics of the correction algorithm, facilitating maintenance personnel to monitor algorithm performance and make necessary parameter adjustments.

[0074] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0075] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An edge collaborative control method of an industrial internet connector, characterized by, The method comprises the following steps: Obtain operation data of a plurality of edge nodes, each operation data containing performance indicators of all connectors in each preset time period; Obtain performance change deviations of each connector in each edge node by comparing the performance indicators of each connector in each edge node with the performance indicators of all connectors in all edge nodes; Obtain performance change values of each connector in each edge node by analyzing the distribution range of the performance indicators of each connector in each edge node and the performance change deviations; Obtain spatial distribution differences of each connector in each edge node by comparing the performance distribution of each connector with all other connectors in each edge node and the performance difference between each connector and its adjacent connector in position; Obtain coordination degrees of each edge node by combining the performance change values and the spatial distribution differences of all connectors in each edge node; Assign control strategies to all edge nodes according to the coordination degrees, and perform edge coordination control based on the edge nodes after the control strategies are assigned. 2.The edge collaborative control method of the industrial internet connector of claim 1, wherein, The performance change deviations are obtained by: Arranging all performance indicator data of each connector in each edge node in time sequence to form each performance sequence, processing all performance indicator data in each performance sequence by using a filtering algorithm to obtain a smooth curve of each performance sequence; Obtaining curvature calculation methods of each smooth curve, calculating integral results of each curvature calculation method in the entire time range; Calculating the average value of the integral results of all connectors in all edge nodes; The performance change deviations are obtained by comparing the integral results of each connector in each edge node with the average value. 3.The edge collaborative control method of the industrial internet connector of claim 2, wherein, The performance change deviations are specifically the difference between the integral results of each connector in each edge node and the average value. 4.The edge collaborative control method of the industrial internet connector of claim 2, wherein, The performance change values are obtained by: Calculating the range of all performance indicator data of each connector in each edge node; The performance change values are positively correlated with the range and the performance change deviations. 5.The edge co-simulation control method of the industrial internet connector according to claim 2, wherein, The spatial distribution differences are obtained by: Calculating the average value of the performance indicator data of each row connector at each collection time, taking the difference between the average value of the performance indicator data of each connector in each edge node and the average value of the performance indicator data of the previous row as a performance value difference, and accumulating the performance value differences of each connector at all collection times to obtain a total performance difference value; Obtaining extreme points of the smooth curve of each connector, arranging the occurrence time of the extreme points of all connectors in each edge node in time sequence to form a time sequence, calculating the time interval of adjacent time points in the time sequence to form a time difference sequence, and calculating the average interval value of the time difference sequence; The spatial distribution differences are obtained by processing the ratio of the total performance difference value and the average interval value by using a normalization function. 6.The edge co-simulation control method of the industrial internet connector of claim 1, wherein, The coordination degrees are obtained by: Calculating the weighted sum value of the performance change value and the spatial distribution difference of each connector in each edge node; The coordination degree is the sum of the weighted sum values of all connectors in each edge node. 7.The edge co-simulation control method of an industrial internet connector of claim 1, wherein, The process of assigning control strategies to all edge nodes according to the coordination degrees is: The initial clustering centers are selected by using a genetic algorithm according to the coordination degrees of the edge nodes, clustering is performed on all the edge nodes based on the initial clustering centers by using a clustering algorithm, high-coordination-degree edge nodes and low-coordination-degree edge nodes are distinguished through the clustering results, different control strategies are assigned to the high-coordination-degree edge nodes and the low-coordination-degree edge nodes, the control strategy of the high-coordination-degree edge nodes is a first preset strategy, and the control strategy of the low-coordination-degree edge nodes is a second preset strategy. 8.The edge collaborative control method of the industrial internet connector of claim 7, wherein, The edge nodes after the control strategies are assigned perform edge coordination control, and the method comprises the following steps of: all the high-coordination-degree edge nodes are divided into a priority execution group and a standard execution group according to a preset proportion, and all the low-coordination-degree edge nodes are divided into the priority execution group and the standard execution group according to the preset proportion; the priority execution group and the standard execution group are used to perform edge coordination control operations. 9.The edge co-simulation control method of the industrial internet connector of claim 1, wherein, The method further comprises the following steps of: historical operation data of each edge node is acquired, and the acquisition process of the performance change bias is corrected based on the historical operation data. 10.The edge co-simulation control method of the industrial internet connector of claim 1, wherein, The method further comprises the following steps of: network topology data of each edge node is acquired, and the acquisition process of the spatial distribution difference is corrected based on the network topology data.

Images